As the algorithms for learning single tasks in restricted environments are improving, new challenges have gained relevance. They include multi-task learning, multimodal sensorimotor learning in open worlds and lifelong adaptation to injury, growth and ageing.

In this workshop we will discuss the developmental processes involved in the emergence of representations of action and perception in humans and artificial agents in continual learning. These processes include action-perception cycle, active perception, continual sensory-motor learning, environmental-driven scaffolding, and intrinsic motivation.

The discussion will be strongly motivated by behavioural and neural data. We hope to provide a discussion friendly environment to connect with research with similar interest regardless of their area of expertise which could include robotics, computer science, psychology, neuroscience, etc. We would also like to devise a roadmap or strategies to develop mathematical and computational models to improve robot performance and/or to attempt to unveil the underlying mechanisms that lead to continual adaptation to changing environment or embodiment and continual learning in open-ended environments.

The primary list of topics covers the following (but not limited to):

• Emergence of representations via continual interaction

• Continual sensory-motor learning

• Action-perception cycle

• Active perception

• Environmental-driven scaffolding

• Intrinsic motivation

• Neural substrates, neural circuits and neural plasticity

• Human and animal behaviour experiments and models

• Reinforcement learning and deep reinforcement learning for life-long learning

• Multisensory robot learning

• Multimodal sensorimotor learning

• Affordance learning

• Prediction learning

Organizers:

• Nicolás Navarro-Guerrero, Aarhus University, Aarhus, Denmark

• Sao Mai Nguyen, IMT Atlantique, France

• Erhan Öztop, Özyeğin University, Turkey

• Junpei Zhong, National Institute of Advanced Industrial Science and Technology (AIST), Japan